You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

model_helper.cc 47 kB

5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
4 years ago
4 years ago
4 years ago
5 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
4 years ago
4 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
5 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
4 years ago
4 years ago
4 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
4 years ago
5 years ago
5 years ago
4 years ago
4 years ago
5 years ago
4 years ago
4 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
4 years ago
4 years ago
5 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
5 years ago
4 years ago
4 years ago
4 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
4 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
4 years ago
4 years ago
4 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
4 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
4 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089
  1. /**
  2. * Copyright 2020 Huawei Technologies Co., Ltd
  3. *
  4. * Licensed under the Apache License, Version 2.0 (the "License");
  5. * you may not use this file except in compliance with the License.
  6. * You may obtain a copy of the License at
  7. *
  8. * http://www.apache.org/licenses/LICENSE-2.0
  9. *
  10. * Unless required by applicable law or agreed to in writing, software
  11. * distributed under the License is distributed on an "AS IS" BASIS,
  12. * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13. * See the License for the specific language governing permissions and
  14. * limitations under the License.
  15. */
  16. #include "framework/common/helper/model_helper.h"
  17. #include "common/model_parser/model_parser.h"
  18. #include "framework/omg/model_tool.h"
  19. #include "framework/omg/version.h"
  20. #include "graph/debug/ge_attr_define.h"
  21. #include "graph/utils/graph_utils.h"
  22. using std::string;
  23. using domi::ModelTaskDef;
  24. namespace {
  25. const int64_t kOriginalOmPartitionNum = 1;
  26. const uint32_t kStatiOmFileModelNum = 1;
  27. }
  28. namespace ge {
  29. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY ModelHelper::~ModelHelper() { (void)ReleaseLocalModelData(); }
  30. Status ModelHelper::SaveModelPartition(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper, ModelPartitionType type,
  31. const uint8_t *data, size_t size, size_t model_index) {
  32. if (size < 1 || size > UINT32_MAX) {
  33. GELOGE(PARAM_INVALID, "[Add][ModelPartition]Failed, partition size %zu invalid", size);
  34. REPORT_INNER_ERROR("E19999", "Add model partition failed, partition size %zu "
  35. "invalid", size);
  36. if (size > UINT32_MAX) {
  37. string item = "item";
  38. if (type == MODEL_DEF) {
  39. item = "model info";
  40. } else if (type == WEIGHTS_DATA) {
  41. item = "weight data";
  42. } else if (type == TASK_INFO) {
  43. item = "task info";
  44. } else if (type == TBE_KERNELS) {
  45. item = "tbe kernels";
  46. } else if (type == CUST_AICPU_KERNELS) {
  47. item = "aicpu kernels";
  48. }
  49. ErrorManager::GetInstance().ATCReportErrMessage("E19023", {"size", "item", "maxsize"},
  50. {std::to_string(size), item, std::to_string(UINT32_MAX)});
  51. }
  52. return PARAM_INVALID;
  53. }
  54. if (data == nullptr) {
  55. GELOGE(PARAM_INVALID, "[Add][ModelPartition]Failed, data is null");
  56. REPORT_INNER_ERROR("E19999", "Add model partition failed, data is null");
  57. return PARAM_INVALID;
  58. }
  59. ModelPartition partition_model;
  60. partition_model.data = const_cast<uint8_t *>(data);
  61. partition_model.size = static_cast<uint32_t>(size);
  62. partition_model.type = type;
  63. if (om_file_save_helper->AddPartition(partition_model, model_index) != SUCCESS) {
  64. GELOGE(PARAM_INVALID, "[Add][ModelPartition]Failed, partition size %zu", size);
  65. REPORT_CALL_ERROR("E19999", "Add model partition failed, partition size %zu", size);
  66. return PARAM_INVALID;
  67. }
  68. return SUCCESS;
  69. }
  70. Status ModelHelper::SaveSizeToModelDef(const GeModelPtr &ge_model) {
  71. vector<int64_t> om_info;
  72. auto ge_model_weight = ge_model->GetWeight();
  73. GELOGD("SaveSizeToModelDef weight_data_size is %zu, %p", ge_model_weight.GetSize(), ge_model_weight.GetData());
  74. om_info.push_back(ge_model_weight.GetSize());
  75. TBEKernelStore tbe_kernel_store = ge_model->GetTBEKernelStore();
  76. GELOGD("SaveSizeToModelDef tbe_kernels_size is %zu", tbe_kernel_store.DataSize());
  77. om_info.push_back(tbe_kernel_store.DataSize());
  78. CustAICPUKernelStore cust_aicpu_kernel_store = ge_model->GetCustAICPUKernelStore();
  79. GELOGD("SaveSizeToModelDef cust aicpu kernels size is %zu", cust_aicpu_kernel_store.DataSize());
  80. om_info.push_back(cust_aicpu_kernel_store.DataSize());
  81. std::shared_ptr<ModelTaskDef> model_task_def = ge_model->GetModelTaskDefPtr();
  82. if (model_task_def == nullptr) {
  83. GELOGD("SaveSizeToModelDef task_info_size is 0.");
  84. om_info.push_back(0);
  85. } else {
  86. size_t partition_task_size = model_task_def->ByteSizeLong();
  87. GELOGD("SaveSizeToModelDef task_info_size is %zu", partition_task_size);
  88. om_info.push_back(partition_task_size);
  89. }
  90. GE_CHK_BOOL_EXEC(ge::AttrUtils::SetListInt(*(ge_model.get()), "om_info_list", om_info),
  91. GELOGE(FAILED, "SetListInt of om_info_list failed.");
  92. return FAILED);
  93. return SUCCESS;
  94. }
  95. Status ModelHelper::SaveModelDef(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper,
  96. const GeModelPtr &ge_model, ge::Buffer &model_buffer, size_t model_index) {
  97. ModelPtr model_tmp = ge::MakeShared<ge::Model>(ge_model->GetName(), ge_model->GetPlatformVersion());
  98. if (model_tmp == nullptr) {
  99. GELOGE(FAILED, "[Creat][Model]Failed, Model %s Ptr", ge_model->GetName().c_str());
  100. REPORT_CALL_ERROR("E19999", "Create Model %s Ptr failed", ge_model->GetName().c_str());
  101. return FAILED;
  102. }
  103. model_tmp->SetGraph(ge_model->GetGraph());
  104. model_tmp->SetVersion(ge_model->GetVersion());
  105. model_tmp->SetAttr(ge_model->MutableAttrMap());
  106. Status ret = SaveSizeToModelDef(ge_model);
  107. if (ret != SUCCESS) {
  108. GELOGE(ret, "[Save][SizeToModelDef]Failed, error_code %u", ret);
  109. REPORT_CALL_ERROR("E19999", "Save SizeToModelDef failed, error_code %u", ret);
  110. return ret;
  111. }
  112. (void)model_tmp->Save(model_buffer);
  113. GELOGD("MODEL_DEF size is %zu", model_buffer.GetSize());
  114. if (model_buffer.GetSize() > 0) {
  115. if (SaveModelPartition(om_file_save_helper, ModelPartitionType::MODEL_DEF, model_buffer.GetData(),
  116. model_buffer.GetSize(), model_index) != SUCCESS) {
  117. GELOGE(PARAM_INVALID, "[Add][ModelPartition]Failed, model_def size %zu, model_index %zu",
  118. model_buffer.GetSize(), model_index);
  119. REPORT_CALL_ERROR("E19999", "Add model graph partititon failed, model_def %zu, "
  120. "model_index %zu", model_buffer.GetSize(), model_index);
  121. return PARAM_INVALID;
  122. }
  123. }
  124. return SUCCESS;
  125. }
  126. Status ModelHelper::SaveModelWeights(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper,
  127. const GeModelPtr &ge_model, size_t model_index) {
  128. auto ge_model_weight = ge_model->GetWeight();
  129. GELOGD("WEIGHTS_DATA size is %zu, %p", ge_model_weight.GetSize(), ge_model_weight.GetData());
  130. // weight is not necessary
  131. if (ge_model_weight.GetSize() > 0) {
  132. GE_CHK_STATUS_RET(SaveModelPartition(om_file_save_helper,
  133. ModelPartitionType::WEIGHTS_DATA,
  134. ge_model_weight.GetData(),
  135. ge_model_weight.GetSize(), model_index), "Add weight partition failed");
  136. }
  137. return SUCCESS;
  138. }
  139. Status ModelHelper::SaveModelTbeKernel(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper,
  140. const GeModelPtr &ge_model, size_t model_index) {
  141. TBEKernelStore tbe_kernel_store = ge_model->GetTBEKernelStore();
  142. GELOGD("TBE_KERNELS size is %zu", tbe_kernel_store.DataSize());
  143. if (tbe_kernel_store.DataSize() > 0) {
  144. GE_CHK_STATUS_RET(
  145. SaveModelPartition(om_file_save_helper, ModelPartitionType::TBE_KERNELS,
  146. ge_model->GetTBEKernelStore().Data(), ge_model->GetTBEKernelStore().DataSize(),
  147. model_index), "Add tbe kernel partition failed");
  148. }
  149. // no need to check value, DATA->NetOutput
  150. (void)tbe_kernel_store.Load(tbe_kernel_store.Data(), tbe_kernel_store.DataSize());
  151. return SUCCESS;
  152. }
  153. Status ModelHelper::SaveModelCustAICPU(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper,
  154. const GeModelPtr &ge_model, size_t model_index) {
  155. CustAICPUKernelStore cust_aicpu_kernel_store = ge_model->GetCustAICPUKernelStore();
  156. GELOGD("cust aicpu kernels size is %zu", cust_aicpu_kernel_store.DataSize());
  157. if (cust_aicpu_kernel_store.DataSize() > 0) {
  158. GE_CHK_STATUS_RET(SaveModelPartition(om_file_save_helper,
  159. ModelPartitionType::CUST_AICPU_KERNELS,
  160. ge_model->GetCustAICPUKernelStore().Data(),
  161. cust_aicpu_kernel_store.DataSize(), model_index),
  162. "Add cust aicpu kernel partition failed");
  163. }
  164. return SUCCESS;
  165. }
  166. Status ModelHelper::SaveModelTaskDef(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper,
  167. const GeModelPtr &ge_model, ge::Buffer &task_buffer, size_t model_index) {
  168. std::shared_ptr<ModelTaskDef> model_task_def = ge_model->GetModelTaskDefPtr();
  169. if (model_task_def == nullptr) {
  170. GELOGE(ACL_ERROR_GE_MEMORY_ALLOCATION, "[Creat][ModelTaskDef]Failed, it is nullptr");
  171. REPORT_CALL_ERROR("E19999", "Creat model task def failed, it is nullptr");
  172. return ACL_ERROR_GE_MEMORY_ALLOCATION;
  173. }
  174. size_t partition_task_size = model_task_def->ByteSizeLong();
  175. GE_IF_BOOL_EXEC(partition_task_size == 0 || partition_task_size > INT_MAX,
  176. GELOGE(FAILED, "[Check][ModelDefSize]Invalid, size %zu!",
  177. partition_task_size);
  178. REPORT_CALL_ERROR("E19999", "Model def size %zu check invalid,",
  179. partition_task_size);
  180. return FAILED);
  181. task_buffer = ge::Buffer(partition_task_size);
  182. if (task_buffer.GetSize() == 0) {
  183. GELOGE(ACL_ERROR_GE_MEMORY_ALLOCATION, "[Allocate][ModelTaskDefBuffer]Failed, "
  184. "model def size %zu", partition_task_size);
  185. REPORT_CALL_ERROR("E19999", "Allocate model task def buffer failed, model def size %zu",
  186. partition_task_size);
  187. return ACL_ERROR_GE_MEMORY_ALLOCATION;
  188. }
  189. (void)model_task_def->SerializePartialToArray(task_buffer.GetData(), static_cast<int>(partition_task_size));
  190. GELOGD("TASK_INFO op_size:%d, stream_num:%u", model_task_def->op().size(), model_task_def->stream_num());
  191. GELOGD("TASK_INFO size is %zu", partition_task_size);
  192. if (SaveModelPartition(om_file_save_helper, ModelPartitionType::TASK_INFO, task_buffer.GetData(),
  193. partition_task_size, model_index) != SUCCESS) {
  194. GELOGE(PARAM_INVALID, "[Add][ModelTaskDefPartition]Failed, model def size %zu, "
  195. "model_index %zu", partition_task_size, model_index);
  196. REPORT_CALL_ERROR("E19999", "Add model task def partition failed, model def size %zu "
  197. "model_index %zu", partition_task_size, model_index);
  198. return PARAM_INVALID;
  199. }
  200. return SUCCESS;
  201. }
  202. Status ModelHelper::SaveModelHeader(std::shared_ptr<OmFileSaveHelper> &om_file_save_helper,
  203. const GeModelPtr &ge_model, size_t model_num) {
  204. // Save target/version to model_header
  205. ModelFileHeader &model_header = om_file_save_helper->GetModelFileHeader();
  206. model_header.platform_type = ge_model->GetPlatformType();
  207. model_header.om_ir_version = ge_model->GetVersion();
  208. model_header.model_num = model_num;
  209. std::string platform_version = ge_model->GetPlatformVersion();
  210. errno_t err;
  211. err = memcpy_s(model_header.platform_version, PLATFORM_VERSION_LEN, platform_version.c_str(),
  212. platform_version.size() + 1);
  213. if (err != EOK) {
  214. GELOGE(ACL_ERROR_GE_MEMORY_ALLOCATION,
  215. "[Save][Model]Failed while allocating memory for platform_version %s, errno %d",
  216. platform_version.c_str(), err);
  217. REPORT_CALL_ERROR("E19999", "ModelHelper save model failed while "
  218. "allocating memory for platform_version %s, errno %d",
  219. platform_version.c_str(), err);
  220. return ACL_ERROR_GE_MEMORY_ALLOCATION;
  221. }
  222. string version = reinterpret_cast<char *>(model_header.platform_version);
  223. GELOGD("Platform version save: %s", version.c_str());
  224. size_t name_size = ge_model->GetName().size();
  225. name_size = name_size > (MODEL_NAME_LENGTH - 1) ? (MODEL_NAME_LENGTH - 1) : name_size;
  226. err = memcpy_s(model_header.name, MODEL_NAME_LENGTH, ge_model->GetName().c_str(), name_size);
  227. if (err != EOK) {
  228. GELOGE(ACL_ERROR_GE_MEMORY_ALLOCATION,
  229. "[Save][Model]Failed while allocating memory for name %s, errno %d",
  230. ge_model->GetName().c_str(), err);
  231. REPORT_CALL_ERROR("E19999", "ModelHelper save model failed while allocating memory "
  232. "for name %s,errno %d", ge_model->GetName().c_str(), err);
  233. return ACL_ERROR_GE_MEMORY_ALLOCATION;
  234. }
  235. string model_name = reinterpret_cast<char *>(model_header.name);
  236. GELOGD("Model name save:%s", model_name.c_str());
  237. return SUCCESS;
  238. }
  239. Status ModelHelper::SaveAllModelPartiton(std::shared_ptr<OmFileSaveHelper>& om_file_save_helper,
  240. const GeModelPtr &ge_model, ge::Buffer &model_buffer,
  241. ge::Buffer &task_buffer, size_t model_index) {
  242. if (SaveModelDef(om_file_save_helper, ge_model, model_buffer, model_index) != SUCCESS) {
  243. GELOGE(FAILED, "[Save][ModelDef]Failed, model index %zu", model_index);
  244. REPORT_INNER_ERROR("E19999", "ModelHelper save model def failed, model index %zu",
  245. model_index);
  246. return FAILED;
  247. }
  248. if (SaveModelWeights(om_file_save_helper, ge_model, model_index) != SUCCESS) {
  249. GELOGE(FAILED, "[Save][ModelWeights]Failed, model index %zu", model_index);
  250. REPORT_INNER_ERROR("E19999","ModelHelper save mode weights failed, mode index %zu",
  251. model_index);
  252. return FAILED;
  253. }
  254. if (SaveModelTbeKernel(om_file_save_helper, ge_model, model_index) != SUCCESS) {
  255. GELOGE(FAILED, "[Save][ModelTbeKernel]Failed, model index %zu", model_index);
  256. REPORT_INNER_ERROR("E19999", "ModelHelper save model tbe kernel failed, model index %zu",
  257. model_index);
  258. return FAILED;
  259. }
  260. if (SaveModelCustAICPU(om_file_save_helper, ge_model, model_index) != SUCCESS) {
  261. GELOGE(FAILED, "[Save][ModelCustAICPU]Failed, model index %zu", model_index);
  262. REPORT_INNER_ERROR("E19999", "ModelHelper save model cust aicpu failed, "
  263. "model index %zu", model_index);
  264. return FAILED;
  265. }
  266. if (SaveModelTaskDef(om_file_save_helper, ge_model, task_buffer, model_index) != SUCCESS) {
  267. GELOGE(FAILED, "[Save][TaskDef]Failed, model index %zu", model_index);
  268. REPORT_INNER_ERROR("E19999", "ModelHelper save task def failed, model index %zu",
  269. model_index);
  270. return FAILED;
  271. }
  272. return SUCCESS;
  273. }
  274. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::SaveToOmModel(const GeModelPtr &ge_model,
  275. const SaveParam &save_param,
  276. const std::string &output_file,
  277. ModelBufferData& model) {
  278. if (output_file.empty()) {
  279. GELOGE(FAILED, "[Save][Model]GraphBuilder SaveModel received invalid file name prefix");
  280. REPORT_INNER_ERROR("E19999", "GraphBuilder SaveModel received invalid file name prefix");
  281. return FAILED;
  282. }
  283. GE_IF_BOOL_EXEC(ge_model == nullptr, GELOGE(FAILED, "Ge_model is nullptr"); return FAILED);
  284. std::shared_ptr<OmFileSaveHelper> om_file_save_helper = ge::MakeShared<OmFileSaveHelper>();
  285. GE_CHECK_NOTNULL(om_file_save_helper);
  286. ge::Buffer model_buffer;
  287. ge::Buffer task_buffer;
  288. auto ret = SaveAllModelPartiton(om_file_save_helper, ge_model, model_buffer, task_buffer);
  289. if (ret != SUCCESS) {
  290. GELOGE(ret, "[Save][AllModelPartition]Failed, error_code %u", ret);
  291. REPORT_INNER_ERROR("E19999", "OmFileSaveHelper save all model partition failed, "
  292. "error_code %u", ret);
  293. return ret;
  294. }
  295. ret = SaveModelHeader(om_file_save_helper, ge_model);
  296. if (ret != SUCCESS) {
  297. GELOGE(ret, "[Save][ModelHeader]Failed, error_code %u", ret);
  298. REPORT_INNER_ERROR("E19999", "OmFileSaveHelper save model header failed, "
  299. "error_code %u", ret);
  300. return ret;
  301. }
  302. ret = om_file_save_helper->SaveModel(save_param, output_file.c_str(), model, is_offline_);
  303. if (ret != SUCCESS) {
  304. GELOGE(FAILED, "[Save][Model]Failed, output file %s", output_file.c_str());
  305. REPORT_INNER_ERROR("E19999", "OmFileSaveHelper save model failed, output file %s",
  306. output_file.c_str());
  307. return ret;
  308. }
  309. return SUCCESS;
  310. }
  311. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::SaveToOmRootModel(
  312. const GeRootModelPtr &ge_root_model,
  313. const SaveParam &save_param,
  314. const std::string &output_file,
  315. ModelBufferData& model,
  316. bool is_unknown_shape) {
  317. GE_CHECK_NOTNULL(ge_root_model);
  318. GE_IF_BOOL_EXEC(ge_root_model == nullptr,
  319. GELOGE(FAILED, "[Check][GERootModel]Ge_root_model is nullptr");
  320. REPORT_INNER_ERROR("E19999", "Ge_root_model check failed, it is nullptr");
  321. return FAILED);
  322. auto &name_to_ge_model = ge_root_model->GetSubgraphInstanceNameToModel();
  323. GE_IF_BOOL_EXEC(name_to_ge_model.empty(),
  324. GELOGE(FAILED, "[Get][SubModel]Ge_root_model has no sub model");
  325. REPORT_INNER_ERROR("E19999", "Ge_root_model has no sub model");
  326. return FAILED);
  327. GE_IF_BOOL_EXEC(output_file.empty(),
  328. GELOGE(FAILED, "[Save][Model]GraphBuilder SaveModel received invalid "
  329. "file name prefix");
  330. REPORT_INNER_ERROR("E19999", "GraphBuilder SaveModel received invalid "
  331. "file name prefix");
  332. return FAILED);
  333. if (!is_unknown_shape) {
  334. auto &model_root = name_to_ge_model.begin()->second;
  335. return SaveToOmModel(model_root, save_param, output_file, model);
  336. }
  337. std::shared_ptr<OmFileSaveHelper> om_file_save_helper = ge::MakeShared<OmFileSaveHelper>();
  338. GE_CHECK_NOTNULL(om_file_save_helper);
  339. auto &first_ge_model = name_to_ge_model.at(ge_root_model->GetRootGraph()->GetName());
  340. // ge root model must be the first to be loaded
  341. vector<string> model_names{ge_root_model->GetRootGraph()->GetName()};
  342. for (auto &item : name_to_ge_model) {
  343. if (item.first != model_names.front()) {
  344. model_names.emplace_back(item.first);
  345. }
  346. }
  347. vector<ge::Buffer> model_buffers(model_names.size());
  348. vector<ge::Buffer> task_buffers(model_names.size());
  349. size_t cur_index = 0;
  350. if (model_names.size() > 1) {
  351. GELOGD("only save first model MODEL_DEF");
  352. if (SaveModelDef(om_file_save_helper, first_ge_model, model_buffers[cur_index], cur_index) != SUCCESS) {
  353. GELOGE(FAILED, "[Save][ModelDef]Failed, cur_index %zu", cur_index);
  354. REPORT_INNER_ERROR("E19999", "Save model def failed, cur_index %zu", cur_index);
  355. return FAILED;
  356. }
  357. ++cur_index;
  358. }
  359. for (; cur_index < model_names.size(); ++cur_index) {
  360. auto model_name = model_names[cur_index];
  361. GELOGD("cur model %s index is %zu", model_name.c_str(), cur_index);
  362. const GeModelPtr &ge_model = name_to_ge_model.at(model_name);
  363. auto ret = SaveAllModelPartiton(om_file_save_helper, ge_model, model_buffers[cur_index],
  364. task_buffers[cur_index], cur_index);
  365. if (ret != SUCCESS) {
  366. GELOGE(INTERNAL_ERROR, "[Save][AllModelPartition]Failed, model name %s, cur_index %zu",
  367. model_name.c_str(), cur_index);
  368. REPORT_INNER_ERROR("E19999", "Save all model %s partition failed, cur_index %zu",
  369. model_name.c_str(), cur_index);
  370. return INTERNAL_ERROR;
  371. }
  372. }
  373. auto ret = SaveModelHeader(om_file_save_helper, first_ge_model, model_names.size());
  374. if (ret != SUCCESS) {
  375. GELOGE(INTERNAL_ERROR, "[Save][ModelHeader]Failed, model name %s",
  376. first_ge_model->GetName().c_str());
  377. REPORT_INNER_ERROR("E19999", "Save model %s header failed", first_ge_model->GetName().c_str());
  378. return INTERNAL_ERROR;
  379. }
  380. ret = om_file_save_helper->SaveRootModel(save_param, output_file.c_str(), model, is_offline_);
  381. if (ret != SUCCESS) {
  382. GELOGE(FAILED, "[Save][Model]OmFileSaveHelper save model return fail, output_file %s",
  383. output_file.c_str());
  384. REPORT_INNER_ERROR("E19999", "OmFileSaveHelper save model return fail, output_file %s",
  385. output_file.c_str());
  386. return FAILED;
  387. }
  388. return SUCCESS;
  389. }
  390. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status
  391. ModelHelper::SaveOriginalGraphToOmModel(const ge::Graph &graph, const std::string &output_file) {
  392. if (output_file.empty()) {
  393. GELOGE(FAILED, "[Save][Model]Received invalid file name prefix, output_file %s",
  394. output_file.c_str());
  395. REPORT_CALL_ERROR("E19999", "Save model received invalid file name prefix, output_file %s",
  396. output_file.c_str());
  397. return FAILED;
  398. }
  399. // Get computegraph from graph
  400. auto compute_graph = ge::GraphUtils::GetComputeGraph(graph);
  401. if (compute_graph == nullptr) {
  402. GELOGE(FAILED, "[Save][Model]Failed for compute_graph null");
  403. REPORT_CALL_ERROR("E19999", "Save model failed for compute_graph null");
  404. return FAILED;
  405. }
  406. GE_DUMP(compute_graph, "OriginalGraph");
  407. // Model
  408. ModelPtr model_ptr = ge::MakeShared<ge::Model>();
  409. GE_CHECK_NOTNULL_EXEC(model_ptr, return MEMALLOC_FAILED);
  410. std::string original_model_name = compute_graph->GetName() + "_original";
  411. model_ptr->SetName(original_model_name);
  412. model_ptr->SetGraph(graph);
  413. model_ptr->SetVersion(static_cast<uint32_t>(OM_PROTO_VERSION));
  414. string framework_version;
  415. Status frame_rt = PlatformVersionManager::GetPlatformVersion(framework_version);
  416. if (frame_rt == SUCCESS) {
  417. uint32_t counter = 0;
  418. string model_framework_version = framework_version + "." + std::to_string(counter);
  419. model_ptr->SetPlatformVersion(model_framework_version);
  420. }
  421. // Model def
  422. ge::Buffer model_buffer;
  423. ge::graphStatus status = model_ptr->Save(model_buffer);
  424. if (status != ge::GRAPH_SUCCESS) {
  425. GELOGE(FAILED, "[Save][Model]Failed for save buffer fail");
  426. REPORT_INNER_ERROR("E19999", "Save model failed for save buffer fail");
  427. return FAILED;
  428. }
  429. std::shared_ptr<OmFileSaveHelper> om_file_save_helper = ge::MakeShared<OmFileSaveHelper>();
  430. GE_CHECK_NOTNULL_EXEC(om_file_save_helper, return MEMALLOC_FAILED);
  431. ModelPartition partition_model;
  432. partition_model.data = model_buffer.GetData();
  433. partition_model.size = static_cast<uint32_t>(model_buffer.GetSize());
  434. partition_model.type = ModelPartitionType::MODEL_DEF;
  435. GELOGI("Original Model type[%u],size[%u]", partition_model.type, partition_model.size);
  436. if (partition_model.data != nullptr && partition_model.size > 0) {
  437. (void)om_file_save_helper->AddPartition(partition_model);
  438. // Condition of AddPartition is established, no need to check value
  439. }
  440. // Save target/version to model_header
  441. ModelFileHeader &model_header = om_file_save_helper->GetModelFileHeader();
  442. model_header.om_ir_version = model_ptr->GetVersion();
  443. model_header.headsize = MODEL_FILE_HEAD_LEN;
  444. std::string platform_version = model_ptr->GetPlatformVersion();
  445. errno_t err = memcpy_s(model_header.platform_version, PLATFORM_VERSION_LEN, platform_version.c_str(),
  446. platform_version.size() + 1);
  447. if (err != EOK) {
  448. GELOGE(FAILED, "[Save][Model]Failed for platform_version %s, errno %d",
  449. platform_version.c_str(), err);
  450. REPORT_CALL_ERROR("E19999", "Save model failed for platform_version %s, errno %d",
  451. platform_version.c_str(), err);
  452. return FAILED;
  453. }
  454. size_t name_size = model_ptr->GetName().size();
  455. name_size = name_size > (MODEL_NAME_LENGTH - 1) ? (MODEL_NAME_LENGTH - 1) : name_size;
  456. err = memcpy_s(model_header.name, MODEL_NAME_LENGTH, model_ptr->GetName().c_str(), name_size);
  457. if (err != EOK) {
  458. GELOGE(FAILED, "[Save][Model]Failed for memory copy %s failed, errno %d",
  459. model_ptr->GetName().c_str(), err);
  460. REPORT_CALL_ERROR("E19999", "Save model failed for memory copy %s failed, errno %d",
  461. model_ptr->GetName().c_str(), err);
  462. return FAILED;
  463. }
  464. ModelBufferData model;
  465. Status ret = om_file_save_helper->SaveModelToFile(output_file.c_str(), model, is_offline_);
  466. return (ret == SUCCESS ? SUCCESS : FAILED);
  467. }
  468. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::LoadModel(const ge::ModelData &model_data) {
  469. if (model_data.model_data == nullptr || model_data.model_len == 0) {
  470. GELOGE(ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID,
  471. "[Load][Model]Model_data is nullptr or model_data_size is 0");
  472. REPORT_INNER_ERROR("E19999", "Load model failed, "
  473. "Model_data is nullptr or model_data_size is 0");
  474. return ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID;
  475. }
  476. if (is_assign_model_) {
  477. GELOGE(ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED, "[Load][Model]Model helper has already loaded!");
  478. REPORT_INNER_ERROR("E19999", "Model helper has already loaded");
  479. return ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED;
  480. }
  481. if (ReleaseLocalModelData() != SUCCESS) {
  482. GELOGE(ACL_ERROR_GE_EXEC_RELEASE_MODEL_DATA, "[Release][ModelData]Failed.");
  483. REPORT_CALL_ERROR("E19999", "Release local model data failed");
  484. return ACL_ERROR_GE_EXEC_RELEASE_MODEL_DATA;
  485. }
  486. Status status = ModelParserBase::ParseModelContent(model_data, model_addr_tmp_, model_len_tmp_);
  487. if (status != SUCCESS) {
  488. GELOGE(ACL_ERROR_GE_PARAM_INVALID, "[Parse][ModelContent]Failed!");
  489. REPORT_CALL_ERROR("E19999", "Parse model content failed");
  490. return ACL_ERROR_GE_PARAM_INVALID;
  491. }
  492. file_header_ = reinterpret_cast<ModelFileHeader *>(model_data.model_data);
  493. OmFileLoadHelper om_load_helper;
  494. status = om_load_helper.Init(model_addr_tmp_, model_len_tmp_);
  495. if (status != SUCCESS) {
  496. GELOGE(status, "[Init][OmLoadHelper]Failed");
  497. REPORT_CALL_ERROR("E19999", "Om_load_helper init failed");
  498. model_addr_tmp_ = nullptr;
  499. return status;
  500. }
  501. auto partition_table = reinterpret_cast<ModelPartitionTable *>(model_addr_tmp_);
  502. if (partition_table->num == kOriginalOmPartitionNum) {
  503. model_addr_tmp_ = nullptr;
  504. GELOGE(ACL_ERROR_GE_PARAM_INVALID, "[Check][OmModel]Error, please use executable om model");
  505. REPORT_CALL_ERROR("E19999", "Om model is error, please use executable om model");
  506. return ACL_ERROR_GE_PARAM_INVALID;
  507. }
  508. // Encrypt model need to del temp model/no encrypt model don't need to del model
  509. model_addr_tmp_ = nullptr;
  510. status = GenerateGeModel(om_load_helper);
  511. if (status != SUCCESS) {
  512. GELOGE(status, "[Generate][GEModel]Failed");
  513. REPORT_CALL_ERROR("E19999", "Generate GE model failed");
  514. return status;
  515. }
  516. GELOGD("in ModelHelper::LoadModel, is_assign_model_ is setted to true!");
  517. is_assign_model_ = true;
  518. return SUCCESS;
  519. }
  520. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::LoadRootModel(const ge::ModelData &model_data) {
  521. if (model_data.model_data == nullptr || model_data.model_len == 0) {
  522. GELOGE(ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID, "[Load][RootModel] "
  523. "Model_data is nullptr or model_data_size is 0");
  524. REPORT_INNER_ERROR("E19999", "Load root model failed, model_data is nullptr or its size is 0");
  525. return ACL_ERROR_GE_EXEC_MODEL_DATA_SIZE_INVALID;
  526. }
  527. if (is_assign_model_) {
  528. GELOGE(ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED,
  529. "[Load][RootModel]Model helper has already loaded!");
  530. REPORT_INNER_ERROR("E19999", "Load root model failed, model helper has already loaded");
  531. return ACL_ERROR_GE_EXEC_LOAD_MODEL_REPEATED;
  532. }
  533. if (ReleaseLocalModelData() != SUCCESS) {
  534. GELOGE(INTERNAL_ERROR, "[Release][ModelData]Failed.");
  535. REPORT_CALL_ERROR("E19999", "Release local root model data failed");
  536. return INTERNAL_ERROR;
  537. }
  538. Status status = ModelParserBase::ParseModelContent(model_data, model_addr_tmp_, model_len_tmp_);
  539. if (status != SUCCESS) {
  540. GELOGE(ACL_ERROR_GE_PARAM_INVALID, "[Parse][RootModelContent]Failed!");
  541. REPORT_INNER_ERROR("E19999", "Parse model content failed");
  542. return ACL_ERROR_GE_PARAM_INVALID;
  543. }
  544. file_header_ = reinterpret_cast<ModelFileHeader *>(model_data.model_data);
  545. //model verison 1.0 file header does not have model_num member
  546. is_unknown_shape_model_ = file_header_->version >= ge::MODEL_VERSION &&
  547. file_header_->model_num > kStatiOmFileModelNum;
  548. GELOGD("cur om model is ge root model or no %d, model version %u", is_unknown_shape_model_, file_header_->version);
  549. OmFileLoadHelper om_load_helper;
  550. if (is_unknown_shape_model_) {
  551. auto model_num = file_header_->model_num;
  552. status = om_load_helper.Init(model_addr_tmp_, model_len_tmp_, model_num);
  553. } else {
  554. status = om_load_helper.Init(model_addr_tmp_, model_len_tmp_);
  555. }
  556. if (status != SUCCESS) {
  557. GELOGE(status, "[Init][OmLoadHelper]Failed");
  558. REPORT_CALL_ERROR("E19999", "Om_load_helper init failed");
  559. model_addr_tmp_ = nullptr;
  560. return status;
  561. }
  562. // Encrypt model need to del temp model/no encrypt model don't need to del model
  563. model_addr_tmp_ = nullptr;
  564. status = GenerateGeRootModel(om_load_helper);
  565. if (status != SUCCESS) {
  566. GELOGE(status, "[Generate][GERootModel]Failed");
  567. REPORT_INNER_ERROR("E19999", "Generate GE root model failed");
  568. return status;
  569. }
  570. GELOGD("in ModelHelper::LoadRootModel, is_assign_model_ is setted to true!");
  571. is_assign_model_ = true;
  572. return SUCCESS;
  573. }
  574. Status ModelHelper::GenerateGeModel(OmFileLoadHelper &om_load_helper) {
  575. model_ = ge::MakeShared<ge::GeModel>();
  576. GE_CHECK_NOTNULL(model_);
  577. Status ret = LoadModelData(om_load_helper);
  578. if (ret != SUCCESS) {
  579. return ACL_ERROR_GE_EXEC_LOAD_MODEL_PARTITION_FAILED;
  580. }
  581. ret = LoadWeights(om_load_helper);
  582. if (ret != SUCCESS) {
  583. return ACL_ERROR_GE_EXEC_LOAD_WEIGHT_PARTITION_FAILED;
  584. }
  585. ret = LoadTask(om_load_helper);
  586. if (ret != SUCCESS) {
  587. return ACL_ERROR_GE_EXEC_LOAD_TASK_PARTITION_FAILED;
  588. }
  589. ret = LoadTBEKernelStore(om_load_helper);
  590. if (ret != SUCCESS) {
  591. return ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED;
  592. }
  593. ret = LoadCustAICPUKernelStore(om_load_helper);
  594. if (ret != SUCCESS) {
  595. return ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED;
  596. }
  597. return SUCCESS;
  598. }
  599. Status ModelHelper::GenerateGeRootModel(OmFileLoadHelper &om_load_helper) {
  600. GELOGD("Begin to generate ge root model");
  601. root_model_ = ge::MakeShared<ge::GeRootModel>();
  602. GE_CHECK_NOTNULL(root_model_);
  603. if (!is_unknown_shape_model_) {
  604. if (GenerateGeModel(om_load_helper) != SUCCESS) {
  605. GELOGE(FAILED, "[Generate][GERootModel]Failed");
  606. REPORT_CALL_ERROR("E19999", "Generate GE root model failed");
  607. return FAILED;
  608. }
  609. GE_CHECK_NOTNULL(model_);
  610. root_model_->SetRootGraph(GraphUtils::GetComputeGraph(model_->GetGraph()));
  611. return SUCCESS;
  612. }
  613. bool is_first_model = true;
  614. for (size_t mode_index = 0; mode_index < file_header_->model_num; ++mode_index) {
  615. GeModelPtr cur_model = ge::MakeShared<ge::GeModel>();
  616. Status ret = LoadModelData(om_load_helper, cur_model, mode_index);
  617. if (ret != SUCCESS) {
  618. return ACL_ERROR_GE_EXEC_LOAD_MODEL_PARTITION_FAILED;
  619. }
  620. if (is_first_model) {
  621. is_first_model = false;
  622. root_model_->SetRootGraph(GraphUtils::GetComputeGraph(cur_model->GetGraph()));
  623. root_model_->SetModelId(cur_model->GetModelId());
  624. model_ = cur_model;
  625. continue;
  626. }
  627. ret = LoadWeights(om_load_helper, cur_model, mode_index);
  628. if (ret != SUCCESS) {
  629. return ACL_ERROR_GE_EXEC_LOAD_WEIGHT_PARTITION_FAILED;
  630. }
  631. ret = LoadTBEKernelStore(om_load_helper, cur_model, mode_index);
  632. if (ret != SUCCESS) {
  633. return ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED;
  634. }
  635. ret = LoadCustAICPUKernelStore(om_load_helper, cur_model, mode_index);
  636. if (ret != SUCCESS) {
  637. return ACL_ERROR_GE_EXEC_LOAD_KERNEL_PARTITION_FAILED;
  638. }
  639. ret = LoadTask(om_load_helper, cur_model, mode_index);
  640. if (ret != SUCCESS) {
  641. return ACL_ERROR_GE_EXEC_LOAD_TASK_PARTITION_FAILED;
  642. }
  643. root_model_->SetSubgraphInstanceNameToModel(cur_model->GetName(), cur_model);
  644. }
  645. return SUCCESS;
  646. }
  647. Status ModelHelper::LoadModelData(OmFileLoadHelper &om_load_helper) {
  648. ModelPartition partition_model_def;
  649. // no need to check value, DATA->NetOutput
  650. om_load_helper.GetModelPartition(ModelPartitionType::MODEL_DEF, partition_model_def);
  651. GELOGD("Model_def partition addr:%p,size:%u", partition_model_def.data, partition_model_def.size);
  652. ge::Model model;
  653. if (ge::Model::Load(partition_model_def.data, partition_model_def.size, model) != SUCCESS) {
  654. GELOGE(INTERNAL_ERROR, "[Load][Model]Failed, model_def partition addr:%p, size:%u",
  655. partition_model_def.data, partition_model_def.size);
  656. REPORT_CALL_ERROR("E19999", "Load model failed, model_def partition addr:%p, size:%u",
  657. partition_model_def.data, partition_model_def.size);
  658. return INTERNAL_ERROR;
  659. }
  660. SetModelToGeModel(model_, model);
  661. return SUCCESS;
  662. }
  663. void ModelHelper::SetModelToGeModel(GeModelPtr &ge_model, Model &model) {
  664. ge_model->SetGraph(model.GetGraph());
  665. ge_model->SetName(model.GetName());
  666. ge_model->SetVersion(model.GetVersion());
  667. ge_model->SetPlatformVersion(model.GetPlatformVersion());
  668. ge_model->SetAttr(model.MutableAttrMap());
  669. }
  670. Status ModelHelper::LoadModelData(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, size_t mode_index) {
  671. ModelPartition partition_model_def;
  672. // no need to check value, DATA->NetOutput
  673. om_load_helper.GetModelPartition(ModelPartitionType::MODEL_DEF, partition_model_def, mode_index);
  674. GELOGD("Model_def partition addr:%p,size:%u", partition_model_def.data, partition_model_def.size);
  675. ge::Model model;
  676. if (ge::Model::Load(partition_model_def.data, partition_model_def.size, model) != SUCCESS) {
  677. GELOGE(INTERNAL_ERROR, "[Load][Model]Failed, model_def partition addr:%p, size:%u",
  678. partition_model_def.data, partition_model_def.size);
  679. REPORT_CALL_ERROR("E19999", "Load model failed, model_def partition addr:%p, size:%u",
  680. partition_model_def.data, partition_model_def.size);
  681. return INTERNAL_ERROR;
  682. }
  683. SetModelToGeModel(cur_model, model);
  684. return SUCCESS;
  685. }
  686. Status ModelHelper::LoadWeights(OmFileLoadHelper &om_load_helper) {
  687. ModelPartition partition;
  688. if (om_load_helper.GetModelPartition(ModelPartitionType::WEIGHTS_DATA, partition) != SUCCESS) {
  689. GELOGE(FAILED, "[Get][ModelWeightPartition]Failed, GetWeight size:%u", partition.size);
  690. REPORT_INNER_ERROR("E19999", "[Get][ModelPartition]Failed, GetWeight size:%u",
  691. partition.size);
  692. return FAILED;
  693. }
  694. ge::Buffer weight = ge::Buffer::CopyFrom(partition.data, partition.size);
  695. model_->SetWeight(weight);
  696. GELOGD("GetWeight size:%u", partition.size);
  697. return SUCCESS;
  698. }
  699. Status ModelHelper::LoadWeights(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, size_t mode_index) {
  700. ModelPartition partition;
  701. if (om_load_helper.GetModelPartition(ModelPartitionType::WEIGHTS_DATA, partition, mode_index) != SUCCESS) {
  702. GELOGE(FAILED, "[Get][ModelPartition]Failed, GetWeight size:%u", partition.size);
  703. REPORT_INNER_ERROR("E19999", "[Get][ModelPartition]Failed, GetWeight size:%u",
  704. partition.size);
  705. return FAILED;
  706. }
  707. ge::Buffer weight = ge::Buffer::CopyFrom(partition.data, partition.size);
  708. cur_model->SetWeight(weight);
  709. GELOGD("GetWeight size:%u", partition.size);
  710. return SUCCESS;
  711. }
  712. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::LoadTask(OmFileLoadHelper &om_load_helper) {
  713. ModelPartition task_partition;
  714. if (om_load_helper.GetModelPartition(ModelPartitionType::TASK_INFO, task_partition) != SUCCESS) {
  715. GELOGE(FAILED, "[Get][ModelTaskPartition]Failed, task_partition size:%u", task_partition.size);
  716. REPORT_INNER_ERROR("E19999", "Get model task partition failed, task_partition size %u",
  717. task_partition.size);
  718. return FAILED;
  719. }
  720. std::shared_ptr<ModelTaskDef> task = ge::MakeShared<ModelTaskDef>();
  721. GE_CHECK_NOTNULL(task);
  722. if (task_partition.size != 0) {
  723. if (!ReadProtoFromArray(task_partition.data, task_partition.size, task.get())) {
  724. GELOGE(INTERNAL_ERROR, "[Read][ProtoFromArray]Failed, task_partition size %u",
  725. task_partition.size);
  726. REPORT_CALL_ERROR("E19999", "Read proto from array failed, task_partition size %u",
  727. task_partition.size);
  728. return INTERNAL_ERROR;
  729. }
  730. GELOGD("TASK_INFO op_size:%d, stream_num:%u", task->op().size(), task->stream_num());
  731. }
  732. model_->SetModelTaskDef(task);
  733. return SUCCESS;
  734. }
  735. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::LoadTask(OmFileLoadHelper &om_load_helper,
  736. GeModelPtr &cur_model,
  737. size_t mode_index) {
  738. ModelPartition task_partition;
  739. if (om_load_helper.GetModelPartition(ModelPartitionType::TASK_INFO, task_partition, mode_index) != SUCCESS) {
  740. GELOGE(FAILED, "Get task model partition failed.");
  741. GELOGE(FAILED, "[Get][ModelTaskPartition]Failed, task_partition size %u, mode_index %zu",
  742. task_partition.size, mode_index);
  743. REPORT_INNER_ERROR("E19999", "Get model task partition failed, "
  744. "task_partition size %u, mode_index %zu", task_partition.size, mode_index);
  745. return FAILED;
  746. }
  747. std::shared_ptr<ModelTaskDef> task = ge::MakeShared<ModelTaskDef>();
  748. GE_CHECK_NOTNULL(task);
  749. if (task_partition.size != 0) {
  750. if (!ReadProtoFromArray(task_partition.data, task_partition.size, task.get())) {
  751. GELOGE(INTERNAL_ERROR, "[Read][ProtoFromArray]Failed, task_partition size %u",
  752. task_partition.size);
  753. REPORT_CALL_ERROR("E19999", "Read proto from array failed, task_partition size %u",
  754. task_partition.size);
  755. return INTERNAL_ERROR;
  756. }
  757. GELOGD("TASK_INFO op_size:%d, stream_num:%u", task->op().size(), task->stream_num());
  758. }
  759. cur_model->SetModelTaskDef(task);
  760. return SUCCESS;
  761. }
  762. Status ModelHelper::LoadTBEKernelStore(OmFileLoadHelper &om_load_helper) {
  763. // Load tbe kernels
  764. ModelPartition partition_kernel_def;
  765. TBEKernelStore kernel_store;
  766. if (om_load_helper.GetModelPartition(ModelPartitionType::TBE_KERNELS, partition_kernel_def) == SUCCESS) {
  767. GELOGD("Kernels partition size:%u", partition_kernel_def.size);
  768. if (kernel_store.Load(partition_kernel_def.data, partition_kernel_def.size)) {
  769. GELOGD("Load tbe kernels success");
  770. } else {
  771. GELOGW("Load tbe kernels failed");
  772. }
  773. }
  774. model_->SetTBEKernelStore(kernel_store);
  775. return SUCCESS;
  776. }
  777. Status ModelHelper::LoadTBEKernelStore(OmFileLoadHelper &om_load_helper, GeModelPtr &cur_model, size_t mode_index) {
  778. // Load tbe kernels
  779. ModelPartition partition_kernel_def;
  780. TBEKernelStore kernel_store;
  781. if (om_load_helper.GetModelPartition(ModelPartitionType::TBE_KERNELS, partition_kernel_def, mode_index) ==
  782. SUCCESS) {
  783. GELOGD("Kernels partition size:%u", partition_kernel_def.size);
  784. if (kernel_store.Load(partition_kernel_def.data, partition_kernel_def.size)) {
  785. GELOGD("Load tbe kernels success");
  786. } else {
  787. GELOGW("Load tbe kernels failed");
  788. }
  789. }
  790. cur_model->SetTBEKernelStore(kernel_store);
  791. return SUCCESS;
  792. }
  793. Status ModelHelper::LoadCustAICPUKernelStore(OmFileLoadHelper &om_load_helper) {
  794. // Load cust aicpu kernels
  795. ModelPartition partition_kernel_def;
  796. CustAICPUKernelStore kernel_store;
  797. if (om_load_helper.GetModelPartition(ModelPartitionType::CUST_AICPU_KERNELS, partition_kernel_def) == SUCCESS) {
  798. GELOGD("Kernels partition size:%u", partition_kernel_def.size);
  799. if (kernel_store.Load(partition_kernel_def.data, partition_kernel_def.size)) {
  800. GELOGD("Load cust aicpu kernels success");
  801. } else {
  802. GELOGW("Load cust aicpu kernels failed");
  803. }
  804. }
  805. model_->SetCustAICPUKernelStore(kernel_store);
  806. return SUCCESS;
  807. }
  808. Status ModelHelper::LoadCustAICPUKernelStore(OmFileLoadHelper &om_load_helper,
  809. GeModelPtr &cur_model, size_t mode_index) {
  810. // Load cust aicpu kernels
  811. ModelPartition partition_kernel_def;
  812. CustAICPUKernelStore kernel_store;
  813. if (om_load_helper.GetModelPartition(ModelPartitionType::CUST_AICPU_KERNELS, partition_kernel_def, mode_index)
  814. == SUCCESS) {
  815. GELOGD("Kernels partition size:%u", partition_kernel_def.size);
  816. if (kernel_store.Load(partition_kernel_def.data, partition_kernel_def.size)) {
  817. GELOGD("Load cust aicpu kernels success");
  818. } else {
  819. GELOGW("Load cust aicpu kernels failed");
  820. }
  821. }
  822. cur_model->SetCustAICPUKernelStore(kernel_store);
  823. return SUCCESS;
  824. }
  825. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY GeModelPtr ModelHelper::GetGeModel() {
  826. if (model_ != nullptr) {
  827. return model_;
  828. }
  829. GELOGD("Model has not been loaded!");
  830. std::shared_ptr<ge::GeModel> out_model = ge::MakeShared<ge::GeModel>();
  831. if (out_model == nullptr) {
  832. return nullptr;
  833. }
  834. return out_model;
  835. }
  836. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY GeRootModelPtr ModelHelper::GetGeRootModel() {
  837. if (root_model_ != nullptr) {
  838. return root_model_;
  839. }
  840. GELOGD("Model has not been loaded!");
  841. std::shared_ptr<ge::GeRootModel> out_model = ge::MakeShared<ge::GeRootModel>();
  842. if (out_model == nullptr) {
  843. return nullptr;
  844. }
  845. return out_model;
  846. }
  847. Status ModelHelper::ReleaseLocalModelData() noexcept {
  848. Status result = SUCCESS;
  849. if (model_addr_tmp_ != nullptr) {
  850. errno_t ret = memset_s(static_cast<void *>(model_addr_tmp_), model_len_tmp_, 0, model_len_tmp_);
  851. if (ret != EOK) {
  852. GELOGE(FAILED, "[Memset][Memory]Failed, error-code %d", ret);
  853. REPORT_CALL_ERROR("E19999", "Failed to memset memory, error-code %d", ret);
  854. result = FAILED;
  855. }
  856. delete[] model_addr_tmp_;
  857. model_addr_tmp_ = nullptr;
  858. model_len_tmp_ = 0;
  859. }
  860. return result;
  861. }
  862. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::GetBaseNameFromFileName(
  863. const string &file_name, string &base_name) {
  864. GELOGD("Get base_name from file, file_name:%s", file_name.c_str());
  865. GE_CHK_BOOL_EXEC_WARN(!file_name.empty(), return FAILED, "File path may not valid, check params --output");
  866. size_t start_position = 0;
  867. // using output as base_name (ignore ".om")
  868. size_t filename_suffixes = 3;
  869. if (file_name.find_last_of('/') != string::npos) {
  870. start_position = file_name.find_last_of('/') + 1;
  871. }
  872. size_t end_position = file_name.length() - filename_suffixes;
  873. base_name = file_name.substr(start_position, end_position - start_position);
  874. GE_CHK_BOOL_EXEC_WARN(!base_name.empty(), return FAILED, "Get base_name failed, check params --output");
  875. return SUCCESS;
  876. }
  877. FMK_FUNC_HOST_VISIBILITY FMK_FUNC_DEV_VISIBILITY Status ModelHelper::GetModelNameFromMergedGraphName(
  878. const string &graph_name, string &model_name) {
  879. GELOGD("Get model_name from graph_name, graph_name:%s", graph_name.c_str());
  880. // this can only be used after merged graph(graph name will be append with "_x", x is index);
  881. GE_CHK_BOOL_EXEC_WARN(!graph_name.empty(), return FAILED, "File path may not valid, check params --output");
  882. size_t start_position = 0;
  883. size_t end_position = graph_name.length();
  884. // using graph as model_name (ignore "_x", x is the index of graph)
  885. if (graph_name.find_last_of('_') != string::npos) {
  886. end_position = graph_name.find_last_of('_');
  887. }
  888. model_name = graph_name.substr(start_position, end_position);
  889. GE_CHK_BOOL_EXEC_WARN(!model_name.empty(), return FAILED, "Get model_name failed, check params --output");
  890. return SUCCESS;
  891. }
  892. Status ModelTool::GetModelInfoFromOm(const char *model_file, ge::proto::ModelDef &model_def, uint32_t &modeldef_size) {
  893. GE_CHECK_NOTNULL(model_file);
  894. ge::ModelData model;
  895. int32_t priority = 0;
  896. Status ret = ModelParserBase::LoadFromFile(model_file, "", priority, model);
  897. if (ret != SUCCESS) {
  898. GELOGE(ret, "[Load][ModelInfo]Failed from file %s, error_code %u", model_file, ret);
  899. REPORT_CALL_ERROR("E19999", "Load model info failed from file %s, error_code %u",
  900. model_file, ret);
  901. return ret;
  902. }
  903. std::function<void()> callback = [&]() {
  904. if (model.model_data != nullptr) {
  905. delete[] reinterpret_cast<char *>(model.model_data);
  906. model.model_data = nullptr;
  907. }
  908. };
  909. GE_MAKE_GUARD(release, callback);
  910. uint8_t *model_data = nullptr;
  911. uint32_t model_len = 0;
  912. ret = ModelParserBase::ParseModelContent(model, model_data, model_len);
  913. if (ret != SUCCESS) {
  914. ErrorManager::GetInstance().ATCReportErrMessage("E10003",
  915. {"parameter", "value", "reason"}, {"om", model_file, "invalid om file"});
  916. GELOGE(ACL_ERROR_GE_PARAM_INVALID,
  917. "[Parse][ModelContent]Failed because of invalid om file. Please check om param.");
  918. return ret;
  919. }
  920. OmFileLoadHelper om_load_helper;
  921. ret = om_load_helper.Init(model_data, model_len);
  922. if (ret != SUCCESS) {
  923. ErrorManager::GetInstance().ATCReportErrMessage("E19021", {"reason"}, {"Om file init failed"});
  924. GELOGE(ge::FAILED, "[Init][OmFile]Failed.");
  925. return ret;
  926. }
  927. ModelPartition ir_part;
  928. ret = om_load_helper.GetModelPartition(MODEL_DEF, ir_part);
  929. if (ret != SUCCESS) {
  930. ErrorManager::GetInstance().ATCReportErrMessage("E19021", {"reason"}, {"Get model part failed"});
  931. GELOGE(ge::FAILED, "[Get][ModelPart]Failed.");
  932. return ret;
  933. }
  934. bool flag = ReadProtoFromArray(ir_part.data, ir_part.size, &model_def);
  935. if (!flag) {
  936. ret = INTERNAL_ERROR;
  937. ErrorManager::GetInstance().ATCReportErrMessage("E19021", {"reason"}, {"ReadProtoFromArray failed"});
  938. GELOGE(ret, "[Read][ProtoFromArray]Failed.");
  939. return ret;
  940. }
  941. modeldef_size = ir_part.size;
  942. return ret;
  943. }
  944. Status ModelTool::GetModelInfoFromPbtxt(const char *model_file, ge::proto::ModelDef &model_def) {
  945. GE_CHECK_NOTNULL(model_file);
  946. ge::ModelData model;
  947. int32_t priority = 0;
  948. Status ret = ModelParserBase::LoadFromFile(model_file, "", priority, model);
  949. auto free_model_data = [](void **ptr) -> void {
  950. if (ptr != nullptr && *ptr != nullptr) {
  951. delete[] reinterpret_cast<char *>(*ptr);
  952. *ptr = nullptr;
  953. }
  954. };
  955. if (ret != SUCCESS) {
  956. free_model_data(&model.model_data);
  957. GELOGE(ret, "[Load][ModelInfo]Failed from file %s, error_code %u", model_file, ret);
  958. REPORT_CALL_ERROR("E19999", "Load model info failed from file %s, error_code %u",
  959. model_file, ret);
  960. return ret;
  961. }
  962. try {
  963. bool flag = google::protobuf::TextFormat::ParseFromString(reinterpret_cast<char *>(model.model_data), &model_def);
  964. if (!flag) {
  965. free_model_data(&model.model_data);
  966. ErrorManager::GetInstance().ATCReportErrMessage("E19021", {"reason"}, {"ParseFromString failed"});
  967. GELOGE(FAILED, "[Parse][ModelInfo]Failed from string.");
  968. return FAILED;
  969. }
  970. free_model_data(&model.model_data);
  971. return SUCCESS;
  972. } catch (google::protobuf::FatalException &e) {
  973. free_model_data(&model.model_data);
  974. ErrorManager::GetInstance().ATCReportErrMessage("E19021", {"reason"}, {"ParseFromString failed, exception message["
  975. + std::string(e.what()) + "]"});
  976. GELOGE(FAILED, "[Parse][ModelInfo]Failed from string, exception message : %s", e.what());
  977. return FAILED;
  978. }
  979. }
  980. } // namespace ge

图引擎模块(GE)是MindSpore的一个子模块,其代码由C++实现,位于前端模块ME和底层硬件之间,起到承接作用。图引擎模块以ME下发的图作为输入,然后进行一系列的深度图优化操作,最后输出一张可以在底层硬件上高效运行的图。GE针对昇腾AI处理器的硬件结构特点,做了特定的优化工作,以此来充分发挥出昇腾AI处理器的强大算力。在进行模型训练/推理时,GE会被自动调用而用户并不感知。GE主要由GE API和GE Core两部分组成,详细的架构图如下所示